@piagent/platform 1.6.1 → 1.8.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +71 -0
- package/README.md +8 -2
- package/README.vi.md +11 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +3 -3
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +2 -2
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +10 -7
- package/docs/context-window-policy.md +3 -3
- package/docs/distribution-standard.md +9 -9
- package/docs/languages.json +2 -2
- package/docs/model-options.md +20 -15
- package/docs/operator-manual-vietnamese.md +12 -12
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +3 -0
- package/docs/quality-benchmark.md +128 -8
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +38 -11
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +1 -1
- package/docs-site/commands.html +1 -1
- package/docs-site/content/en/runtime.html +1 -1
- package/docs-site/content/en/whats-new.html +23 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +23 -0
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +1 -1
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +1 -1
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +2 -2
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -1
- package/docs-site/en/workflows.html +1 -1
- package/docs-site/index.html +1 -1
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +3 -3
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -1
- package/docs-site/workflows.html +1 -1
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +9 -5
- package/evals/fs5-pilot-protocol.v1.json +1 -1
- package/evals/fs5-pilot-protocol.v4.json +4 -4
- package/evals/fs5-pilot-protocol.v5.json +6 -5
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +2 -2
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +107 -107
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +63 -50
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +69 -8
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +162 -45
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +269 -3
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +79 -4
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +144 -128
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +211 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +30 -2
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +395 -33
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +3 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +264 -70
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +139 -152
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +11 -6
- package/packages/piagent-core/extensions/criterion-context-pack.js +96 -4
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/guard-types.ts +5 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +197 -155
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +74 -38
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +60 -8
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +1 -0
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +76 -83
- package/packages/piagent-core/runtime/hooks/input-hook.ts +92 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +50 -18
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +53 -20
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +49 -8
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +61 -15
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/model-route-policy.ts +33 -13
- package/packages/piagent-core/runtime/model/model-route-types.ts +10 -6
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +12 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +11 -2
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +192 -57
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +27 -39
- package/packages/piagent-core/runtime/recovery/resume-state.ts +8 -5
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +12 -8
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/policy-tools.ts +5 -3
- package/packages/piagent-core/runtime/registration/session-commands.ts +9 -7
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +27 -11
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +5 -2
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +23 -23
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -22
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +29 -35
- package/packages/piagent-core/runtime/session/system-prompt.ts +3 -3
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +3 -2
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +10 -3
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +11 -11
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +44 -7
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +221 -242
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-B4XehTCU.js → MarkdownMessage-DMU7Xe_O.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-Cui9AAV_.js +176 -0
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +10 -2
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +6 -7
- package/packages/piagent-webui/extension/session-stream-adapter.ts +60 -11
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +141 -19
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +14 -6
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +114 -114
- package/packages/piagent-webui/server/core-inspection-provider.ts +73 -14
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +65 -12
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +32 -1
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +30 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +217 -388
- package/scripts/benchmark-runner-finalization.mjs +140 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-model-scope.sh +6 -2
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/install-global.sh +2 -2
- package/scripts/model-route-evaluation-core.ts +1 -0
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-model-route.mjs +1 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/setup.sh +2 -2
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +37 -56
- package/scripts/verify-release-identity.mjs +5 -3
- package/templates/global/settings.json +5 -1
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -0,0 +1,483 @@
|
|
|
1
|
+
import { createHash, randomUUID } from "node:crypto";
|
|
2
|
+
import path from "node:path";
|
|
3
|
+
import { types } from "node:util";
|
|
4
|
+
|
|
5
|
+
import { SCOPED_CODEX_TURN_OBSERVATION_VERSION, SCOPED_MCP_LIMITS,
|
|
6
|
+
SCOPED_MCP_METADATA_CONTRACT, SCOPED_TOOL_DEFINITIONS
|
|
7
|
+
} from "./benchmark-scoped-frozen-qualification.mjs";
|
|
8
|
+
|
|
9
|
+
const hash = value => createHash("sha256").update(value).digest("hex");
|
|
10
|
+
const plain = value => Boolean(value && typeof value === "object" && !types.isProxy(value)
|
|
11
|
+
&& [null, Object.prototype].includes(Object.getPrototypeOf(value)));
|
|
12
|
+
function deepFreeze(value) {
|
|
13
|
+
if (!value || typeof value !== "object" || Object.isFrozen(value)) return value;
|
|
14
|
+
for (const child of Object.values(value)) deepFreeze(child);
|
|
15
|
+
return Object.freeze(value);
|
|
16
|
+
}
|
|
17
|
+
const brokerFail = code => { throw Object.assign(new Error(code), { brokerCode: code }); };
|
|
18
|
+
const brokerRequire = (value, code) => { if (!value) brokerFail(code); };
|
|
19
|
+
const brokerWellFormed = value => typeof value === "string" && value.isWellFormed();
|
|
20
|
+
function brokerFields(value, names) {
|
|
21
|
+
brokerRequire(plain(value), "invalid-object");
|
|
22
|
+
const descriptors = Object.getOwnPropertyDescriptors(value);
|
|
23
|
+
brokerRequire(Reflect.ownKeys(descriptors).length === names.length && names.every(name =>
|
|
24
|
+
descriptors[name]?.enumerable && Object.hasOwn(descriptors[name], "value")), "invalid-fields");
|
|
25
|
+
}
|
|
26
|
+
const MCP_VERSION = "2025-06-18";
|
|
27
|
+
const MCP_PROCESS_SHUTDOWN = "piagent-mcp-process-shutdown";
|
|
28
|
+
const MCP_META_ID = /^[A-Za-z0-9_.:-]{1,160}$/;
|
|
29
|
+
const MCP_GIT_COMMIT = /^(?:[a-f0-9]{40}|[a-f0-9]{64})$/;
|
|
30
|
+
const MCP_TURN_FIELDS = SCOPED_MCP_METADATA_CONTRACT.turnFields;
|
|
31
|
+
const MCP_TURN_REQUIRED_FIELDS = Object.freeze(MCP_TURN_FIELDS.filter(key => key !== "reasoning_effort"));
|
|
32
|
+
const MCP_CALL_META_FIELDS = SCOPED_MCP_METADATA_CONTRACT.callFields;
|
|
33
|
+
const rpcId = value => Number.isSafeInteger(value) || typeof value === "string" && value.length > 0
|
|
34
|
+
&& value.length <= 160 && brokerWellFormed(value);
|
|
35
|
+
const rpcError = (requestId, code, message) => ({ jsonrpc: "2.0", id: requestId,
|
|
36
|
+
error: { code, message } });
|
|
37
|
+
|
|
38
|
+
// The broker kernel registers a private journal append capability here. This
|
|
39
|
+
// keeps protocol transport and child loading outside the auditable kernel while
|
|
40
|
+
// preserving an in-process-only capability: it is never serialized to Codex.
|
|
41
|
+
const brokerPorts = new WeakMap();
|
|
42
|
+
|
|
43
|
+
export function registerScopedBrokerTransport(broker, append, seal) {
|
|
44
|
+
brokerRequire(broker && typeof broker === "object" && typeof broker.status === "function"
|
|
45
|
+
&& typeof append === "function" && typeof seal === "function" && !brokerPorts.has(broker),
|
|
46
|
+
"mcp-broker-binding");
|
|
47
|
+
brokerPorts.set(broker, { append, seal, used: false });
|
|
48
|
+
}
|
|
49
|
+
|
|
50
|
+
function mcpMetaExact(value, allowed, required = []) {
|
|
51
|
+
brokerRequire(plain(value), "mcp-invalid-metadata");
|
|
52
|
+
const keys = Object.keys(value);
|
|
53
|
+
brokerRequire(keys.every(key => allowed.includes(key)) && required.every(key => Object.hasOwn(value, key)),
|
|
54
|
+
"mcp-invalid-metadata");
|
|
55
|
+
}
|
|
56
|
+
const mcpMetaId = value => typeof value === "string" && MCP_META_ID.test(value) && brokerWellFormed(value);
|
|
57
|
+
const mcpMetaText = (value, maxBytes = 2048) => typeof value === "string" && value.length > 0
|
|
58
|
+
&& Buffer.byteLength(value) <= maxBytes && brokerWellFormed(value);
|
|
59
|
+
function validateMcpWorkspaces(value) {
|
|
60
|
+
brokerRequire(plain(value), "mcp-invalid-metadata");
|
|
61
|
+
const entries = Object.entries(value);
|
|
62
|
+
brokerRequire(entries.length > 0 && entries.length <= 16, "mcp-invalid-metadata");
|
|
63
|
+
for (const [workspace, state] of entries) {
|
|
64
|
+
brokerRequire(mcpMetaText(workspace) && path.isAbsolute(workspace) && path.normalize(workspace) === workspace,
|
|
65
|
+
"mcp-invalid-metadata");
|
|
66
|
+
mcpMetaExact(state, ["latest_git_commit_hash", "has_changes"], ["has_changes"]);
|
|
67
|
+
brokerRequire(typeof state.has_changes === "boolean"
|
|
68
|
+
&& (!Object.hasOwn(state, "latest_git_commit_hash")
|
|
69
|
+
|| typeof state.latest_git_commit_hash === "string"
|
|
70
|
+
&& MCP_GIT_COMMIT.test(state.latest_git_commit_hash)), "mcp-invalid-metadata");
|
|
71
|
+
}
|
|
72
|
+
}
|
|
73
|
+
function validateMcpTurnMetadata(value) {
|
|
74
|
+
mcpMetaExact(value, MCP_TURN_FIELDS, MCP_TURN_REQUIRED_FIELDS);
|
|
75
|
+
for (const key of ["session_id", "thread_id", "turn_id", "model"]) {
|
|
76
|
+
brokerRequire(mcpMetaId(value[key]), "mcp-invalid-metadata");
|
|
77
|
+
}
|
|
78
|
+
brokerRequire(Number.isSafeInteger(value.turn_started_at_unix_ms) && value.turn_started_at_unix_ms >= 0
|
|
79
|
+
&& value.thread_source === "user" && mcpMetaText(value.sandbox, 80)
|
|
80
|
+
&& value.sandbox_mode === "workspace-write"
|
|
81
|
+
&& typeof value.node_repl_disabled === "boolean" && typeof value.auto_review_enabled === "boolean"
|
|
82
|
+
&& typeof value.node_repl_auto_review_required === "boolean", "mcp-invalid-metadata");
|
|
83
|
+
validateMcpWorkspaces(value.workspaces);
|
|
84
|
+
if (Object.hasOwn(value, "reasoning_effort")) brokerRequire(typeof value.reasoning_effort === "string"
|
|
85
|
+
&& ["none", "minimal", "low", "medium", "high", "xhigh", "max", "ultra"].includes(value.reasoning_effort),
|
|
86
|
+
"mcp-invalid-metadata");
|
|
87
|
+
}
|
|
88
|
+
function validateMcpSpecialPath(value) {
|
|
89
|
+
mcpMetaExact(value, ["kind", "subpath"]);
|
|
90
|
+
brokerRequire(["root", "minimal", "project_roots", "tmpdir", "slash_tmp"].includes(value.kind),
|
|
91
|
+
"mcp-invalid-metadata");
|
|
92
|
+
if (value.kind === "project_roots") {
|
|
93
|
+
if (Object.hasOwn(value, "subpath")) brokerRequire(mcpMetaText(value.subpath, 1024), "mcp-invalid-metadata");
|
|
94
|
+
} else brokerRequire(!Object.hasOwn(value, "subpath"), "mcp-invalid-metadata");
|
|
95
|
+
}
|
|
96
|
+
function validateMcpSandboxPath(value) {
|
|
97
|
+
brokerRequire(plain(value) && ["path", "glob_pattern", "special"].includes(value.type),
|
|
98
|
+
"mcp-invalid-metadata");
|
|
99
|
+
if (value.type === "path") {
|
|
100
|
+
mcpMetaExact(value, ["type", "path"], ["type", "path"]);
|
|
101
|
+
brokerRequire(mcpMetaText(value.path), "mcp-invalid-metadata");
|
|
102
|
+
} else if (value.type === "glob_pattern") {
|
|
103
|
+
mcpMetaExact(value, ["type", "pattern"], ["type", "pattern"]);
|
|
104
|
+
brokerRequire(mcpMetaText(value.pattern), "mcp-invalid-metadata");
|
|
105
|
+
} else {
|
|
106
|
+
mcpMetaExact(value, ["type", "value"], ["type", "value"]); validateMcpSpecialPath(value.value);
|
|
107
|
+
}
|
|
108
|
+
}
|
|
109
|
+
function validateMcpFileSystemPermissions(value) {
|
|
110
|
+
brokerRequire(plain(value) && ["restricted", "unrestricted"].includes(value.type), "mcp-invalid-metadata");
|
|
111
|
+
if (value.type === "unrestricted") return mcpMetaExact(value, ["type"], ["type"]);
|
|
112
|
+
mcpMetaExact(value, ["type", "entries", "glob_scan_max_depth"], ["type", "entries"]);
|
|
113
|
+
brokerRequire(Array.isArray(value.entries) && value.entries.length <= 64, "mcp-invalid-metadata");
|
|
114
|
+
if (Object.hasOwn(value, "glob_scan_max_depth")) brokerRequire(Number.isSafeInteger(value.glob_scan_max_depth)
|
|
115
|
+
&& value.glob_scan_max_depth > 0 && value.glob_scan_max_depth <= 64, "mcp-invalid-metadata");
|
|
116
|
+
for (const entry of value.entries) {
|
|
117
|
+
mcpMetaExact(entry, ["path", "access", "missing_path_behavior"], ["path", "access"]);
|
|
118
|
+
validateMcpSandboxPath(entry.path);
|
|
119
|
+
brokerRequire(["read", "write", "deny"].includes(entry.access), "mcp-invalid-metadata");
|
|
120
|
+
if (Object.hasOwn(entry, "missing_path_behavior")) brokerRequire(entry.missing_path_behavior === "skip",
|
|
121
|
+
"mcp-invalid-metadata");
|
|
122
|
+
}
|
|
123
|
+
}
|
|
124
|
+
function validateMcpPermissionProfile(value) {
|
|
125
|
+
brokerRequire(plain(value) && ["managed", "disabled", "external"].includes(value.type),
|
|
126
|
+
"mcp-invalid-metadata");
|
|
127
|
+
if (value.type === "disabled") return mcpMetaExact(value, ["type"], ["type"]);
|
|
128
|
+
if (value.type === "external") {
|
|
129
|
+
mcpMetaExact(value, ["type", "network"], ["type", "network"]);
|
|
130
|
+
} else {
|
|
131
|
+
mcpMetaExact(value, ["type", "file_system", "network"], ["type", "file_system", "network"]);
|
|
132
|
+
validateMcpFileSystemPermissions(value.file_system);
|
|
133
|
+
}
|
|
134
|
+
brokerRequire(["restricted", "enabled"].includes(value.network), "mcp-invalid-metadata");
|
|
135
|
+
}
|
|
136
|
+
function validateMcpSandboxState(value) {
|
|
137
|
+
mcpMetaExact(value, ["permissionProfile", "codexLinuxSandboxExe", "sandboxCwd", "useLegacyLandlock"],
|
|
138
|
+
["permissionProfile", "codexLinuxSandboxExe", "sandboxCwd", "useLegacyLandlock"]);
|
|
139
|
+
validateMcpPermissionProfile(value.permissionProfile);
|
|
140
|
+
brokerRequire((value.codexLinuxSandboxExe === null || typeof value.codexLinuxSandboxExe === "string"
|
|
141
|
+
&& path.isAbsolute(value.codexLinuxSandboxExe) && path.normalize(value.codexLinuxSandboxExe) === value.codexLinuxSandboxExe
|
|
142
|
+
&& Buffer.byteLength(value.codexLinuxSandboxExe) <= 2048 && brokerWellFormed(value.codexLinuxSandboxExe))
|
|
143
|
+
&& mcpMetaText(value.sandboxCwd) && value.sandboxCwd.startsWith("file:")
|
|
144
|
+
&& typeof value.useLegacyLandlock === "boolean", "mcp-invalid-metadata");
|
|
145
|
+
}
|
|
146
|
+
function mcpBusinessParams(params, method) {
|
|
147
|
+
if (!params || !Object.hasOwn(params, "_meta")) return { business: params, observation: null };
|
|
148
|
+
const metadata = params._meta;
|
|
149
|
+
const allowed = method === "tools/call" ? MCP_CALL_META_FIELDS : ["progressToken"];
|
|
150
|
+
mcpMetaExact(metadata, allowed);
|
|
151
|
+
brokerRequire(Buffer.byteLength(JSON.stringify(metadata)) <= SCOPED_MCP_METADATA_CONTRACT.maxBytes,
|
|
152
|
+
"mcp-invalid-metadata");
|
|
153
|
+
if (Object.hasOwn(metadata, "progressToken")) {
|
|
154
|
+
const token = metadata.progressToken;
|
|
155
|
+
brokerRequire(Number.isSafeInteger(token) || brokerWellFormed(token)
|
|
156
|
+
&& Buffer.byteLength(token) <= 160, "mcp-invalid-metadata");
|
|
157
|
+
}
|
|
158
|
+
const codexFields = ["callId", "threadId", "itemId", "x-codex-turn-metadata", "codex/sandbox-state-meta"];
|
|
159
|
+
const hasCodexFields = codexFields.some(key => Object.hasOwn(metadata, key));
|
|
160
|
+
let observation = null;
|
|
161
|
+
if (hasCodexFields) {
|
|
162
|
+
brokerRequire(method === "tools/call" && ["callId", "threadId", "itemId", "x-codex-turn-metadata"]
|
|
163
|
+
.every(key => Object.hasOwn(metadata, key)) && [metadata.callId, metadata.threadId, metadata.itemId].every(mcpMetaId),
|
|
164
|
+
"mcp-invalid-metadata");
|
|
165
|
+
validateMcpTurnMetadata(metadata["x-codex-turn-metadata"]);
|
|
166
|
+
brokerRequire(metadata.threadId === metadata["x-codex-turn-metadata"].thread_id,
|
|
167
|
+
"mcp-invalid-metadata");
|
|
168
|
+
if (Object.hasOwn(metadata, "codex/sandbox-state-meta")) validateMcpSandboxState(metadata["codex/sandbox-state-meta"]);
|
|
169
|
+
const turn = metadata["x-codex-turn-metadata"];
|
|
170
|
+
observation = deepFreeze({ version: SCOPED_CODEX_TURN_OBSERVATION_VERSION,
|
|
171
|
+
contractId: SCOPED_MCP_METADATA_CONTRACT.id, authority: "none",
|
|
172
|
+
metadataSha256: hash(JSON.stringify(metadata)), callId: metadata.callId,
|
|
173
|
+
threadId: metadata.threadId, itemId: metadata.itemId, sessionId: turn.session_id,
|
|
174
|
+
turnId: turn.turn_id, turnStartedAtUnixMs: turn.turn_started_at_unix_ms,
|
|
175
|
+
model: turn.model, reasoningEffort: turn.reasoning_effort ?? null,
|
|
176
|
+
sandboxMode: turn.sandbox_mode,
|
|
177
|
+
sandboxStateSha256: Object.hasOwn(metadata, "codex/sandbox-state-meta")
|
|
178
|
+
? hash(JSON.stringify(metadata["codex/sandbox-state-meta"])) : null });
|
|
179
|
+
}
|
|
180
|
+
const business = { ...params }; delete business._meta;
|
|
181
|
+
return { business, observation };
|
|
182
|
+
}
|
|
183
|
+
|
|
184
|
+
function parseMcpFrame(bytes) {
|
|
185
|
+
let text, value;
|
|
186
|
+
try {
|
|
187
|
+
text = new TextDecoder("utf-8", { fatal: true }).decode(bytes);
|
|
188
|
+
value = JSON.parse(text, (key, item) => {
|
|
189
|
+
brokerRequire(brokerWellFormed(key) && (typeof item !== "string" || brokerWellFormed(item))
|
|
190
|
+
&& (typeof item !== "number" || Number.isFinite(item)), "mcp-invalid-json");
|
|
191
|
+
return item;
|
|
192
|
+
});
|
|
193
|
+
} catch { brokerFail("mcp-invalid-json"); }
|
|
194
|
+
const stack = [];
|
|
195
|
+
for (const match of text.matchAll(/"(?:\\.|[^"\\])*"|[{}\[\],:]/g)) {
|
|
196
|
+
const token = match[0], current = stack.at(-1);
|
|
197
|
+
if (token === "{") stack.push({ keys: new Set(), key: true });
|
|
198
|
+
else if (token === "[") stack.push({ keys: null });
|
|
199
|
+
else if (token === "}" || token === "]") stack.pop();
|
|
200
|
+
else if (current?.keys && token === ",") current.key = true;
|
|
201
|
+
else if (current?.keys && token === ":") current.key = false;
|
|
202
|
+
else if (current?.keys && current.key && token.startsWith('"')) {
|
|
203
|
+
const key = JSON.parse(token);
|
|
204
|
+
brokerRequire(!current.keys.has(key), "mcp-duplicate-json-key"); current.keys.add(key);
|
|
205
|
+
}
|
|
206
|
+
brokerRequire(stack.length <= 16, "mcp-json-depth");
|
|
207
|
+
}
|
|
208
|
+
brokerRequire(value && typeof value === "object" && !Array.isArray(value) && value.jsonrpc === "2.0"
|
|
209
|
+
&& typeof value.method === "string" && value.method.length <= 160
|
|
210
|
+
&& Object.keys(value).every(key => ["jsonrpc", "id", "method", "params"].includes(key))
|
|
211
|
+
&& (!Object.hasOwn(value, "params") || value.params && typeof value.params === "object"
|
|
212
|
+
&& !Array.isArray(value.params))
|
|
213
|
+
&& (!Object.hasOwn(value, "id") || rpcId(value.id)), "mcp-invalid-request");
|
|
214
|
+
return value;
|
|
215
|
+
}
|
|
216
|
+
|
|
217
|
+
/** Bounded byte-stream MCP adapter. It receives a broker object and opaque
|
|
218
|
+
* journal capability, never host verifier plans, source, Docker, or signing key. */
|
|
219
|
+
export async function runScopedBrokerMcp({ broker, input, output, nonce, signal,
|
|
220
|
+
limits = SCOPED_MCP_LIMITS }) {
|
|
221
|
+
const port = brokerPorts.get(broker);
|
|
222
|
+
brokerRequire(port && !port.used && !broker.status().ended && nonce === broker.identity.nonce,
|
|
223
|
+
"mcp-broker-binding");
|
|
224
|
+
brokerFields(limits, Object.keys(SCOPED_MCP_LIMITS));
|
|
225
|
+
for (const key of Object.keys(limits)) brokerRequire(Number.isSafeInteger(limits[key])
|
|
226
|
+
&& limits[key] > 0 && limits[key] <= SCOPED_MCP_LIMITS[key], "mcp-invalid-limit");
|
|
227
|
+
limits = { ...limits };
|
|
228
|
+
brokerRequire(input !== output && typeof input?.[Symbol.asyncIterator] === "function"
|
|
229
|
+
&& typeof input.destroy === "function" && typeof output?.write === "function"
|
|
230
|
+
&& typeof output.end === "function" && typeof output.destroy === "function"
|
|
231
|
+
&& !input.readableObjectMode && !output.writableObjectMode, "mcp-byte-streams-required");
|
|
232
|
+
port.used = true;
|
|
233
|
+
const transportId = randomUUID(), seenIds = new Set(), deadline = Date.now() + limits.timeoutMs;
|
|
234
|
+
let phase = "new", frames = 0, outputFrames = 0, inputBytes = 0, outputBytes = 0;
|
|
235
|
+
let buffer = Buffer.alloc(0), failure = null, recorded = true, rejectStop, inflight = null;
|
|
236
|
+
let queuedSends = 0, terminalized = false, outputEnded = false;
|
|
237
|
+
const decidedFrames = new Set(), writtenOutputFrames = new Set();
|
|
238
|
+
const pending = new Set();
|
|
239
|
+
const stopped = new Promise((_, reject) => { rejectStop = reject; }); stopped.catch(() => {});
|
|
240
|
+
const stop = reason => {
|
|
241
|
+
if (!failure) { failure = reason; rejectStop(Object.assign(new Error(reason), { brokerCode: reason })); }
|
|
242
|
+
};
|
|
243
|
+
const check = () => {
|
|
244
|
+
if (Date.now() >= deadline) stop("mcp-timeout");
|
|
245
|
+
brokerRequire(!failure, failure);
|
|
246
|
+
};
|
|
247
|
+
const onError = () => { if (!terminalized) stop("mcp-stream-error"); };
|
|
248
|
+
const timer = setTimeout(() => stop("mcp-timeout"), limits.timeoutMs);
|
|
249
|
+
input.on("error", onError); output.on("error", onError);
|
|
250
|
+
input.once("close", () => input.off("error", onError));
|
|
251
|
+
output.once("close", () => output.off("error", onError));
|
|
252
|
+
const audit = event => {
|
|
253
|
+
try { port.append({ transportId, ...event }); }
|
|
254
|
+
catch (error) { recorded = false; throw error; }
|
|
255
|
+
};
|
|
256
|
+
// A deferred call has two promise-finally bookkeeping steps after its reply,
|
|
257
|
+
// transport-write, and transport-decision are already durable. Codex can
|
|
258
|
+
// close stdin and terminate the process group in that narrow gap. Treat the
|
|
259
|
+
// transport as settled once every accepted frame has a durable decision,
|
|
260
|
+
// every emitted frame has a durable write record, and the broker has no
|
|
261
|
+
// actual verifier work left. Stale local promise references carry no more
|
|
262
|
+
// journal or output work and must not prevent normal shutdown sealing.
|
|
263
|
+
const transportSettled = () => phase === "ready" && buffer.length === 0
|
|
264
|
+
&& decidedFrames.size === frames && writtenOutputFrames.size === outputFrames
|
|
265
|
+
&& !broker.status().inflightVerification
|
|
266
|
+
&& (typeof input.readableLength !== "number" || input.readableLength === 0);
|
|
267
|
+
const quiescent = () => transportSettled() && !inflight && pending.size === 0 && queuedSends === 0;
|
|
268
|
+
const closeOutput = () => {
|
|
269
|
+
if (outputEnded) return;
|
|
270
|
+
outputEnded = true; output.end();
|
|
271
|
+
};
|
|
272
|
+
const sealTransport = () => {
|
|
273
|
+
if (terminalized) return true;
|
|
274
|
+
try {
|
|
275
|
+
port.seal({ transportId, type: "transport-end", complete: !failure, reason: failure,
|
|
276
|
+
frames, outputFrames, inputBytes, outputBytes, phase });
|
|
277
|
+
} catch { recorded = false; failure ??= "journal-unavailable"; }
|
|
278
|
+
terminalized = true;
|
|
279
|
+
return broker.status().ended;
|
|
280
|
+
};
|
|
281
|
+
const finalizeSettled = () => {
|
|
282
|
+
if (terminalized) return true;
|
|
283
|
+
if (!transportSettled()) return false;
|
|
284
|
+
sealTransport();
|
|
285
|
+
try { closeOutput(); }
|
|
286
|
+
catch { recorded = false; failure ??= "mcp-stream-error"; }
|
|
287
|
+
return true;
|
|
288
|
+
};
|
|
289
|
+
const finalizeQuiescent = () => quiescent() && finalizeSettled();
|
|
290
|
+
const shutdownError = () => Object.assign(new Error(MCP_PROCESS_SHUTDOWN),
|
|
291
|
+
{ brokerCode: MCP_PROCESS_SHUTDOWN });
|
|
292
|
+
const onInputEnd = () => { if (!failure) finalizeSettled(); };
|
|
293
|
+
const onAbort = () => {
|
|
294
|
+
if (signal?.reason === MCP_PROCESS_SHUTDOWN && transportSettled()) {
|
|
295
|
+
finalizeSettled(); rejectStop(shutdownError());
|
|
296
|
+
} else stop("mcp-cancelled");
|
|
297
|
+
// Promise.race stops the main loop; destroying the stream also releases
|
|
298
|
+
// the losing iterator.next() so no stdin handle can hold the child open
|
|
299
|
+
// until Codex's two-second SIGKILL escalation.
|
|
300
|
+
if (!input.destroyed) input.destroy(signal?.reason === MCP_PROCESS_SHUTDOWN
|
|
301
|
+
? shutdownError() : Object.assign(new Error("mcp-cancelled"), { brokerCode: "mcp-cancelled" }));
|
|
302
|
+
};
|
|
303
|
+
signal?.addEventListener("abort", onAbort, { once: true });
|
|
304
|
+
input.once("end", onInputEnd);
|
|
305
|
+
const wait = promise => Promise.race([promise, stopped]);
|
|
306
|
+
async function send(reply) {
|
|
307
|
+
check(); const bytes = Buffer.from(JSON.stringify(reply) + "\n");
|
|
308
|
+
brokerRequire(bytes.length <= limits.frameBytes && outputBytes + bytes.length <= limits.outputBytes,
|
|
309
|
+
"mcp-output-limit");
|
|
310
|
+
const frame = ++outputFrames, sha256 = hash(bytes); outputBytes += bytes.length;
|
|
311
|
+
audit({ type: "transport-frame", direction: "out", frame, sha256, bytes: bytes.length,
|
|
312
|
+
state: "prepared" });
|
|
313
|
+
await wait(new Promise((resolve, reject) =>
|
|
314
|
+
output.write(bytes, error => error ? reject(error) : resolve())));
|
|
315
|
+
check(); audit({ type: "transport-write", frame, sha256, state: "write-callback" });
|
|
316
|
+
writtenOutputFrames.add(frame);
|
|
317
|
+
}
|
|
318
|
+
let sendChain = Promise.resolve();
|
|
319
|
+
const queueSend = reply => {
|
|
320
|
+
queuedSends++;
|
|
321
|
+
const sending = sendChain.then(() => send(reply)).finally(() => { queuedSends--; });
|
|
322
|
+
sendChain = sending.catch(() => {}); return sending;
|
|
323
|
+
};
|
|
324
|
+
function dispatch(request) {
|
|
325
|
+
const requestId = Object.hasOwn(request, "id") ? request.id : null;
|
|
326
|
+
const error = (code, message) => requestId === null ? null : rpcError(requestId, code, message);
|
|
327
|
+
if (requestId !== null) {
|
|
328
|
+
brokerRequire(!seenIds.has(requestId), "mcp-duplicate-id"); seenIds.add(requestId);
|
|
329
|
+
}
|
|
330
|
+
if (request.method === "notifications/initialized") {
|
|
331
|
+
brokerRequire(requestId === null && phase === "initializing"
|
|
332
|
+
&& (!request.params || Object.keys(request.params).length === 0), "mcp-ready-order");
|
|
333
|
+
phase = "ready"; return null;
|
|
334
|
+
}
|
|
335
|
+
if (request.method === "notifications/cancelled") {
|
|
336
|
+
brokerRequire(phase === "ready" && requestId === null, "mcp-cancel-shape");
|
|
337
|
+
const target = rpcId(request.params?.requestId) ? request.params.requestId : null;
|
|
338
|
+
if (inflight && target === inflight.requestId) {
|
|
339
|
+
if (!broker.status().cancelled) broker.cancel();
|
|
340
|
+
audit({ type: "transport-cancellation", requestId: target, outcome: "signalled",
|
|
341
|
+
reason: "inflight-verification" });
|
|
342
|
+
} else audit({ type: "transport-cancellation", requestId: target, outcome: "ignored",
|
|
343
|
+
reason: "no-matching-inflight-request" });
|
|
344
|
+
return null;
|
|
345
|
+
}
|
|
346
|
+
brokerRequire(requestId !== null, "mcp-unsupported-notification");
|
|
347
|
+
const decoded = mcpBusinessParams(request.params, request.method);
|
|
348
|
+
request = { ...request, params: decoded.business };
|
|
349
|
+
if (decoded.observation) audit({ type: "transport-observation", requestId,
|
|
350
|
+
observation: decoded.observation });
|
|
351
|
+
if (request.method === "initialize") {
|
|
352
|
+
brokerRequire(phase === "new", "mcp-initialize-order");
|
|
353
|
+
brokerFields(request.params, ["protocolVersion", "capabilities", "clientInfo"]);
|
|
354
|
+
const info = request.params.clientInfo;
|
|
355
|
+
brokerRequire(typeof request.params.protocolVersion === "string"
|
|
356
|
+
&& /^\d{4}-\d{2}-\d{2}$/.test(request.params.protocolVersion)
|
|
357
|
+
&& request.params.capabilities && typeof request.params.capabilities === "object"
|
|
358
|
+
&& !Array.isArray(request.params.capabilities) && info && typeof info.name === "string"
|
|
359
|
+
&& info.name.length <= 160 && typeof info.version === "string" && info.version.length <= 160
|
|
360
|
+
&& Object.keys(info).every(key => ["name", "version", "title"].includes(key)),
|
|
361
|
+
"mcp-initialize-parameters");
|
|
362
|
+
phase = "initializing";
|
|
363
|
+
return { jsonrpc: "2.0", id: requestId, result: { protocolVersion: MCP_VERSION,
|
|
364
|
+
capabilities: { tools: {} }, serverInfo: { name: "piagent-scoped-broker", version: "1" } } };
|
|
365
|
+
}
|
|
366
|
+
brokerRequire(phase === "ready", "mcp-not-ready");
|
|
367
|
+
if (inflight) return error(-32001, "request-inflight");
|
|
368
|
+
if (request.method === "ping" || request.method === "tools/list") {
|
|
369
|
+
if (request.params !== undefined) brokerFields(request.params, []);
|
|
370
|
+
return { jsonrpc: "2.0", id: requestId,
|
|
371
|
+
result: request.method === "ping" ? {} : { tools: SCOPED_TOOL_DEFINITIONS } };
|
|
372
|
+
}
|
|
373
|
+
if (request.method !== "tools/call") return error(-32601, "method-not-found");
|
|
374
|
+
try { brokerFields(request.params, ["name", "arguments"]); }
|
|
375
|
+
catch {
|
|
376
|
+
try { broker.invoke(request.params?.name, null, nonce); }
|
|
377
|
+
catch { /* The kernel records this invalid call. */ }
|
|
378
|
+
brokerRequire(!broker.status().blocked, "journal-unavailable");
|
|
379
|
+
return error(-32602, "invalid-tool-parameters");
|
|
380
|
+
}
|
|
381
|
+
const present = result => {
|
|
382
|
+
brokerRequire(!result.lines || result.lines.length <= 256, "mcp-output-record-limit");
|
|
383
|
+
return { jsonrpc: "2.0", id: requestId, result: { content: [{ type: "text",
|
|
384
|
+
text: JSON.stringify(result) }], structuredContent: result, isError: false } };
|
|
385
|
+
};
|
|
386
|
+
const denied = errorValue => {
|
|
387
|
+
brokerRequire(!broker.status().blocked && errorValue.brokerCode !== "mcp-output-record-limit",
|
|
388
|
+
errorValue.brokerCode ?? "mcp-kernel-error");
|
|
389
|
+
if (["tool-denied", "invalid-fields", "invalid-object", "invalid-argument", "invalid-tool",
|
|
390
|
+
"invalid-id"].includes(errorValue.brokerCode)) return error(-32602, "invalid-tool-or-arguments");
|
|
391
|
+
return { jsonrpc: "2.0", id: requestId, result: { isError: true,
|
|
392
|
+
content: [{ type: "text", text: errorValue.brokerCode ?? "operation-error" }] } };
|
|
393
|
+
};
|
|
394
|
+
if (request.params.name === "scoped_verify" && broker.status().verificationAvailable) {
|
|
395
|
+
return { deferred: broker.invokeAsync(request.params.name, request.params.arguments, nonce)
|
|
396
|
+
.then(present, denied), requestId, method: request.method };
|
|
397
|
+
}
|
|
398
|
+
try { return present(broker.invoke(request.params.name, request.params.arguments, nonce)); }
|
|
399
|
+
catch (errorValue) { return denied(errorValue); }
|
|
400
|
+
}
|
|
401
|
+
try {
|
|
402
|
+
audit({ type: "transport-begin", protocolVersion: MCP_VERSION,
|
|
403
|
+
toolsSha256: hash(JSON.stringify(SCOPED_TOOL_DEFINITIONS)), limits });
|
|
404
|
+
if (signal?.aborted) stop("mcp-cancelled");
|
|
405
|
+
const iterator = input[Symbol.asyncIterator]();
|
|
406
|
+
for (;;) {
|
|
407
|
+
check(); const next = await wait(iterator.next()); if (next.done) break;
|
|
408
|
+
brokerRequire(Buffer.isBuffer(next.value), "mcp-nonbyte-input"); inputBytes += next.value.length;
|
|
409
|
+
brokerRequire(inputBytes <= limits.inputBytes, "mcp-input-limit");
|
|
410
|
+
buffer = Buffer.concat([buffer, next.value]);
|
|
411
|
+
for (let index; (index = buffer.indexOf(10)) >= 0;) {
|
|
412
|
+
check(); const frame = buffer.subarray(0, index + 1); buffer = buffer.subarray(index + 1); frames++;
|
|
413
|
+
audit({ type: "transport-frame", direction: "in", frame: frames, sha256: hash(frame),
|
|
414
|
+
bytes: frame.length });
|
|
415
|
+
brokerRequire(frame.length <= limits.frameBytes && frames <= limits.frames, "mcp-frame-limit");
|
|
416
|
+
let request, reply;
|
|
417
|
+
try { request = parseMcpFrame(frame.subarray(0, -1)); reply = dispatch(request); }
|
|
418
|
+
catch (error) {
|
|
419
|
+
audit({ type: "transport-decision", frame: frames, requestId: request?.id ?? null,
|
|
420
|
+
method: request?.method ?? null, outcome: "denied",
|
|
421
|
+
reason: error.brokerCode ?? "mcp-invalid-request" });
|
|
422
|
+
decidedFrames.add(frames);
|
|
423
|
+
if (recorded && !broker.status().blocked && (!request || Object.hasOwn(request, "id"))
|
|
424
|
+
&& error.brokerCode !== "mcp-cancelled") {
|
|
425
|
+
const code = !request ? (error.brokerCode === "mcp-invalid-json" ? -32700 : -32600)
|
|
426
|
+
: error.brokerCode === "mcp-not-ready" ? -32002 : -32602;
|
|
427
|
+
await send(rpcError(request?.id ?? null, code, "request-rejected"));
|
|
428
|
+
}
|
|
429
|
+
throw error;
|
|
430
|
+
}
|
|
431
|
+
if (reply?.deferred) {
|
|
432
|
+
const active = { requestId: reply.requestId, method: reply.method, frame: frames, promise: null };
|
|
433
|
+
inflight = active;
|
|
434
|
+
active.promise = reply.deferred.then(async deferredReply => {
|
|
435
|
+
await queueSend(deferredReply);
|
|
436
|
+
audit({ type: "transport-decision", frame: active.frame, requestId: active.requestId,
|
|
437
|
+
method: active.method, outcome: deferredReply?.error || deferredReply?.result?.isError
|
|
438
|
+
? "denied" : "processed" });
|
|
439
|
+
decidedFrames.add(active.frame);
|
|
440
|
+
}).catch(error => { stop(error.brokerCode ?? "mcp-kernel-error"); }).finally(() => {
|
|
441
|
+
if (inflight === active) inflight = null;
|
|
442
|
+
});
|
|
443
|
+
pending.add(active.promise); active.promise.finally(() => pending.delete(active.promise));
|
|
444
|
+
} else {
|
|
445
|
+
if (reply) await queueSend(reply);
|
|
446
|
+
audit({ type: "transport-decision", frame: frames, requestId: request.id ?? null,
|
|
447
|
+
method: request.method, outcome: reply?.error || reply?.result?.isError ? "denied" : "processed" });
|
|
448
|
+
decidedFrames.add(frames);
|
|
449
|
+
}
|
|
450
|
+
}
|
|
451
|
+
brokerRequire(buffer.length <= limits.frameBytes, "mcp-frame-limit");
|
|
452
|
+
}
|
|
453
|
+
brokerRequire(buffer.length === 0 && phase === "ready", "mcp-incomplete-stream");
|
|
454
|
+
if (pending.size) await wait(Promise.all([...pending]));
|
|
455
|
+
await wait(sendChain);
|
|
456
|
+
brokerRequire(finalizeQuiescent(), "mcp-incomplete-stream");
|
|
457
|
+
check();
|
|
458
|
+
} catch (error) {
|
|
459
|
+
if (error.brokerCode !== MCP_PROCESS_SHUTDOWN) stop(error.brokerCode ?? "mcp-stream-error");
|
|
460
|
+
}
|
|
461
|
+
finally {
|
|
462
|
+
clearTimeout(timer); signal?.removeEventListener("abort", onAbort); input.off("end", onInputEnd);
|
|
463
|
+
try {
|
|
464
|
+
if (buffer.length) audit({ type: "transport-tail", bytes: buffer.length, sha256: hash(buffer) });
|
|
465
|
+
if (failure && !broker.status().blocked && !broker.status().cancelled) broker.cancel();
|
|
466
|
+
if (broker.status().inflightVerification) {
|
|
467
|
+
broker.cancelVerification(); await broker.reconcileVerification();
|
|
468
|
+
}
|
|
469
|
+
if (pending.size) await Promise.allSettled([...pending]);
|
|
470
|
+
await sendChain;
|
|
471
|
+
if (!terminalized) sealTransport();
|
|
472
|
+
} catch { recorded = false; failure ??= "journal-unavailable"; }
|
|
473
|
+
terminalized = true; input.destroy();
|
|
474
|
+
if (failure) output.destroy();
|
|
475
|
+
else {
|
|
476
|
+
try { closeOutput(); }
|
|
477
|
+
catch { recorded = false; failure ??= "mcp-stream-error"; }
|
|
478
|
+
}
|
|
479
|
+
}
|
|
480
|
+
const status = broker.status();
|
|
481
|
+
return Object.freeze({ complete: !failure && recorded, recorded, reason: failure, frames, outputFrames,
|
|
482
|
+
inputBytes, outputBytes, g0Qualified: false, verificationAvailable: status.verificationAvailable });
|
|
483
|
+
}
|